What machine learning is
Traditional software follows instructions a person wrote. Machine learning flips that. You give a program many examples, and it works out the rules itself. Show it thousands of labeled photos of cats and dogs, and it learns what separates them without anyone describing whiskers or ears.
The result of that learning is a model: a file of numbers, called parameters, that turns an input into an output. Training is the process of adjusting those numbers until the outputs are good.
How it works in plain terms
Think of a student doing practice problems with an answer key. The student guesses, checks the key, sees how wrong the guess was, and adjusts. Repeat a few million times and the guesses get good.
In a computer, the "adjusting" is usually gradient descent: a method that nudges every parameter a little in the direction that reduces the error. The examples are the training data. A model that memorizes its examples instead of learning the pattern is overfitting, and it fails on anything new.
The main styles
- Supervised learning: the examples come with answers, such as emails labeled spam or not spam.
- Unsupervised learning: no answers are given, and the model finds structure on its own, such as grouping similar customers.
- Reinforcement learning: the model tries actions and gets rewards or penalties, the way a game-playing program improves by playing.
Large language models mix these. They are first trained to predict the next piece of text from huge amounts of writing, then tuned with human feedback and rewards. See pretraining and post-training.
Why it matters to you
Almost every AI tool you use, from spam filters to chatbots to image generators, is machine learning underneath. Knowing the idea explains the behavior you see. A model is only as good as its examples, so it can repeat the gaps and biases in its training data. It finds patterns rather than looking up facts, which is one reason it can state something false with confidence. That failure is called hallucination.
An example
A bank wants to flag fraud. Instead of writing a thousand "if the amount is over X and the country is Y" rules, it feeds a model years of transactions marked fraudulent or fine. The model learns combinations of signals no analyst would have listed, and it keeps improving as new examples arrive.
How it relates to neighboring terms
Machine learning is the broad field. Deep learning is a branch that uses many-layered neural networks. A large language model is one kind of deep learning model, built for text.